On Mar 12, 2010, at 9:28 PM, Chris Trahey wrote: > The old class is still there, think of it as if the inserted (overloading) > class extends the base (overloaded) class and any classes the were extending > the base now extend the inserted class. So as far as the runtime, it's > run-of-the-meill inheritance. Methods that are not re-implimented in the > inserted class are called in the original class, etc. > > It could be implemented either at the time a class is loaded (when we see > 'overloads' keyword) or perhaps in a function call: > overload_class('Library_Class', 'My_LibClass_Overload'); > > As for conflicts where multiple overloads are attempted, they could be in > sequence, such that you'd end up having an inheritance chain like this: > > called_class <-- second_overload <-- first_overload <-- Library_Class <-- > etc.
In Objective-C, one did this in the older days by calling NSObject's -poseAsClass: method, which would replace the class you sent it to with the class you specified. Nowadays that approach is strongly discouraged in favor of method swizzling, a technique by which you replace a specific class' implementation of a single method with your own (which, with a minor bit of extra work, can call through to the original). Internally, it's done by replacing the function pointer in the class' method table with a new one and returning the old one. Zend's implementation is slightly more twisted than libobjc's, amusingly enough, but it's still doable. This is the typical way for a dynamic language to solve the problem you've described. In PHP I believe the necessary work is already implemented in runkit via the function runkit_method_redefine(). You need only install that, or refer to it to make an extension of your own. If you want this functionality in core, file a feature request. (I'm personally in favor of making a lot of runkit's dynamic class tweaking abilities core). As far as I understand your issue, this technique would solve it cleanly. > On Fri, Mar 12, 2010 at 8:11 PM, Etienne Kneuss <col...@php.net> wrote: > >> Hi, >> >> On Sat, Mar 13, 2010 at 2:50 AM, Chris Trahey <christra...@gmail.com> >> wrote: >>> Perhaps a new concept in class-based OO programming, I'm not sure. >>> >>> Depending on your perspective you could call it ancestor overloading (or >>> upstream overloading) or class underloading. >>> >>> >>> We are increasingly developing with the aid of frameworks & libraries. In >>> fact, this idea came from my current project using the Zend Framework. >>> >>> These libraries, while greatly extensible, are also fairly >> self-extending. >>> That is, they include many classes that extend many classes, which is >> great. >>> >>> As consumers of these libraries, we can extend the classes and consume >> the >>> API however we please, but there is one sticking point. >>> >>> We cannot change classes that many other classes extend without extending >> or >>> changing each child class and then making sure that our code uses the new >>> class. >>> >>> >>> For a concrete example, I was working with the Zend_Form_Element >> subclasses, >>> and I realized that I wanted to change some of the default behavior (in >>> Zend_Form_Element). >>> >>> - at this point I will assume the reader understands why I wouldn't want >> to >>> just start changing the Zend library files - >>> >>> There are many subclasses of Zend_Form_Element. If you want to change the >>> default behavior for all of them, you have 3 choices currently: >>> >>> 1. Directly edit the Zend_Form_Element file in the library, -bad for >> updates >>> & other projects that use the library >>> >>> 2. subclass Zend_Form_Element and change declaration of the descendants >> to >>> extend new class - same problems >>> >>> 3. extend each child class and implement those subclasses in your app >> code >>> -very tedious and TONS of repeated code, breaks consistency of API for >>> developers. >>> >>> >>> There could be a better way, if we could insert a class into the family >>> tree. >>> >>> And that's the heart of this idea, so I'll repeat it: >>> >>> * insert a class into the family tree * >>> >>> >>> Image we do it using an alternative keyword to "extends", such as >>> "overloads". >>> >>> >>> Example: >>> >>> >>> class Library_Class { } >>> >>> class Library_Subclass extends Library_Class {} >>> >>> and then: >>> >>> class My_LibClass_Overload overloads Library_Class{} >>> >>> >>> Now new instances of Library_Subclass actually extend >> My_LibClass_Overload, >>> which "extends" Library_Class. The developer would then code >>> My_LibClass_Overload as if it were declared like this: >>> >>> class Library_Class {} >>> >>> class My_LibClass_Overload extends Library_Class {} >>> >>> class Library_Subclass extends My_LibClass_Overload {} >>> >>> >>> But indeed the declaration of Library_Subclass would *not* have to >> change. >>> >>> >>> This way developers could "extend" default functionality and have >> *existing* >>> library classes pick up the new functionality without redeclaring >> anything >>> in the library. >>> >>> Downstream classes would still override any methods that they redeclare. >> If >>> you wanted to have end-point classes in the library have different >> behavior, >>> you would overload them instead, such as >>> >>> class My_LibSubclass_Overload overloads Lib_Subclass {} >>> >>> >>> The benefit is that the application code can still consume "standard" >>> classes, such as Library_Subclass and not need to know or care about the >>> extended functionality. >>> >>> >>> Going back to my concrete example, my code could then still use >>> Zend_Form_Element_Text, but benefit from the modifications I added, >> without >>> me having to touch the library code. >>> >>> >>> I hope I've explained clearly what this could look like. I'm a younger >>> developer, so forgive me if I'm rough on the terminology -perhaps >>> overload/underload is not the best word for this functionality. Also, I'm >>> not sure if there are other class-based OO languages that allow this kind >> of >>> behavior... Prototypal languages perhaps, as is the case with javascript >> and >>> the Obj.prototype which (combined with anonymous functions) allows you to >>> extend the "base" functionality of other objects that "extend" it. >> >> Even though it might look appealing from a framework user perspective, >> it looks fishy from a language design perspective. It sounds like >> you're trying to fix a framework design lack by a language trick. >> >> For the fishy part: what happens to the old class? what about static >> method calls on that old class? What if two classes overwrites the >> same class? Basically it would mean there is no way to know at compile >> time which class new Foo; is supposed to instantiate. >> >>> >>> >>> Thank you for your comments and thoughts! >>> >>> >>> Chris Trahey >>> >>> Web Applications Developer >>> >>> Database Administrator >>> >>> CSISD [Technology] >>> >>> >>> footnote: I sent this message from a different address and it did not >> show >>> up. I tested sending to internals-h...@lists.php.net and did not get a >>> response -so I assume there is an outgoing issue on my other server's >> side. >>> Forgive me if this message shows up again. >>> >> >> >> >> -- >> Etienne Kneuss >> http://www.colder.ch >> -- Gwynne -- PHP Internals - PHP Runtime Development Mailing List To unsubscribe, visit: http://www.php.net/unsub.php